État académique
Thèse soutenue le 2015-02-19
Sujet: Modèles descriptifs de textures à base d'éléments pour l'interprétation d'images médicales
Direction de thèse:
Encadrement de thèse:
Ellipse bleue: doctorant, ellipse jaune: docteur, rectangle vert: permanent, rectangle jaune: HDR. Trait vert: encadrant de thèse, trait bleu: directeur de thèse, pointillé: jury d'évaluation à mi-parcours ou jury de thèse.
Productions scientifiques
Comparison of the Spatial Organization in Colorectal Tumors using Second-Order Statistics and Functional ANOVA
We propose an automatic method to quantitatively describe the spatial organization governing populations of biological objects, such as cells, which exist in stationary histology images. This quantification is of prime importance when striving to compare different tumoral models in order to evaluate potential therapies. We compare two animal models of colorectal cancer. Our approach is based on the topographic map to automatically extract the location of the relevant biological objects. We describe their spatial organization along a continuous range of scales using second-order statistics. Using a functional analysis of variance test, we show that there are significant differences in these statistics depending on cancer model, and on the day after tumor implant.
Image and Signal Processing and Analysis (ISPA), 2013 8th International Symposium on Image and Signal Processing and Analysis (ISPA), 2013 8th International Symposium onconference proceeding 2013
Glaucoma Detection based on Local Binary Patterns in Fundus Photographs
Glaucoma, a group of diseases that lead to optic neuropathy, is one of the most common reasons for blindness worldwide. Glaucoma rarely causes symptoms until the later stages of the disease. Early detection of glaucoma is very important to prevent visual loss since optic nerve damages cannot be reversed. To detect glaucoma, purely data-driven techniques have advantages, especially when the disease characteristics are complex and when precise image-based measurements are difficult to obtain. In this paper, we present our preliminary study for glaucoma detection using an automatic method based on local texture features extracted from fundus photographs. It implements the completed modeling of Local Binary Patterns to capture representative texture features from the whole image. A local region is represented by three operators: its central pixel (LBPC) and its local differences as two complementary components, the sign (which is the classical LBP) and the magnitude (LBPM). An image texture is finally described by both the distribution of LBP and the joint-distribution of LBPM and LBPC. Our images are then classified using a nearest-neighbor method with a leave-one-out validation strategy. On a sample set of 41 fundus images (13 glaucomatous, 28 non-glaucomatous), our method achieves 95:1% success rate with a specificity of 92:3% and a sensitivity of 96:4%. This study proposes a reproducible glaucoma detection process that could be used in a low-priced medical screening, thus avoiding the inter-experts variability issue. © (2014) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Proc. SPIE 9035, Medical Imaging 2014: Computer-Aided Diagnosis SPIE Medical Imagingconference proceeding 2014-03-18
Grading cancer from liver histology images using inter and intra region spatial relations
Histology image analysis is widely used is cancer studies since this modality preserves the tissue structure. In this paper, we propose a framework to grade metastatic liver histology images based on the spatial organization inter and intra regions. After detecting the presence of metastases, we first decompose the image into regions corresponding to the tissue types (sane, cancerous, vessels and gaps). A sample of each type is further decomposed into the contained biological objects (nuclei, stroma, gaps). The spatial relations between all the pairs of regions and objects are measured using a Force Histogram Decomposition. A specimen is described using a Bag of Words model aggregating the features measured on all its randomly acquired images. The grading is finally made using a Naive Bayes Classifier. Experiments on a 23 mice dataset with CT26 intrasplenic tumors highlight the relevance of the spatial relations with a correct grading rate of 78.95%.
Lecture Notes in Computer Sciences International Conference on Image Analysis and Recognitionconference proceeding 2014
Thèse: Analyse statistique des populations pour l'interprétation d'images histologiques
Soutenance: 2015-02-19